Project Summary/Abstract: Epilepsy is a debilitating neurological disease affecting 50 million people worldwide. While current management of epilepsy depends on continuous medication or invasive surgery, a deeper understanding of circuit-level pathology and an ability to predict seizures reliably will yield more focused therapeutic approaches. We will induce spontaneous unilateral focal seizures in rats by focal injection of kainic acid into the hippocampus. We will continuously record intracranial EEG, LFP, and video data from these rats for a period of 120 days. Relevant features will then be computed, including signal correlations, coherence, energy, entropies, and phase synchronization, as well as features detected by a sparse autoencoder. We will build train predictive classifiers using the first 90 days of data input to the following algorithms: (1) linear regression, (2) random forest, (3) support vector machines, and (4) neural networks. Performance will be assessed by ROC computed for the last 30 days of data. Next, we will apply single pulse electrical stimuli periodically to a new cohort of continuously monitored rats with unilateral hippocampal focal epilepsy. We will analyze the responses to this single pulse electrical stimulus on intracranial EEG and depth electrodes to determine whether seizures can be interrupted, predicted, then preempted by stimulus. Machine learning approaches coupled with active electrical stimulation will render algorithms that identify changes in the electrographic response to stimulus, which predict subsequent seizures with high sensitivity and specificity. We will leverage the fine spatiotemporal resolution offered by optogenetic techniques to study the nature of seizure initiation at the circuit level in a closed-loop system. We will probe the hippocampal circuit considered a generator in focal epilepsy: pyramidal neurons expressing CamK2a. To target each of these elements, we will induce expression of a halorhodopsin driven by the CamK2a promoter via a lentivirus delivered stereotactically to the hippocampal CA3. We will then stimulate the halorhodopsin with 589 nm light pulses, delivered during pre-ictal activity identified by the predictive algorithms and stimulus response. This closed-loop system will be uniquely informative by altering brain states preemptively rather than during the actual seizure as done in previous studies. Our study will ultimately offer new insights into the mechanisms underlying seizure initiation, and may help improve diagnostics and therapeutic approaches for patients suffering from focal epilepsy.